All outputs are real tool calls to the Equibles MCP server from 21 October 2026, trimmed.
Reading one earnings call is easy. Telling what changed since the last one is the hard part: the slides look alike, every quarter is "solid", and the shift hides in which topics got airtime and how carefully management spoke.
Here's how I get an AI agent to make that comparison with numbers, using Bank of America's last two calls and one cross-bank check.
In Claude, open Customize → Connectors → Discover, search for Equibles, click Connect to Claude and Authorize, then switch it on in a chat from + → Connectors. In ChatGPT: Plugins, search Equibles, Install plugin. In Claude Code, claude mcp add --transport http equibles https://mcp.equibles.com/mcp and then /mcp to sign in. The free plan covers 100 requests a day; other clients are in the setup docs.
The prompt: "Using Equibles, compare the tone and themes of Bank of America's last two earnings calls." The agent calls GetEarningsCallToneAndThemes for BAC with limit=2:
| FY2026 Q2 (14 Jul) | FY2026 Q3 (14 Oct) | |
|---|---|---|
| Tone (-100 cautious to +100 bullish) | 78 | 57 |
| Hedging (0 plain to 100 qualified) | 25 | 34 |
And the Q3 themes, with each theme's tone and how often it came up:
| Q3 theme | Tone | Mentions |
|---|---|---|
| Markets and investment banking activity | risk | 64 |
| Loan and deposit growth | positive | 32 |
| Net interest income and yield growth | positive | 31 |
| AI and operating leverage | positive | 27 |
| Wealth management growth | positive | 17 |
| Credit quality | positive | 16 |
| Regulatory settlement and remediation | risk | 16 |
| Consumer banking and rewards | positive | 13 |
In Q2, seven of eight themes were positive and the eighth neutral. In Q3, two carry a risk tone. A tone of 57 is still positive; what the pair tells you is direction: less upbeat, more qualified.
Two calls can mislead, so widen the window. "Show BAC's call themes over the last four quarters." GetEarningsCallThemeHistory with calls=4 returns a theme-by-quarter grid. Trimmed (a dash means not raised; + positive, ! risk):
| Theme | FY25 Q4 | FY26 Q1 | FY26 Q2 | FY26 Q3 | Status |
|---|---|---|---|---|---|
| Net interest income and yield growth | 80 + | 37 + | 48 + | 31 + | recurring |
| Markets and investment banking activity | 24 + | 55 + | - | 64 ! | recurring |
| AI and operating leverage | - | - | 37 + | 27 + | recurring |
| Credit quality | - | - | - | 16 + | new |
| Regulatory settlement and remediation | - | - | - | 16 ! | new |
| Consumer banking and rewards | - | - | - | 13 + | new |
| Expense discipline and operating efficiency | 33 + | 37 + | - | - | dropped after Q1 |
Read the pairing before trusting a "new" or "dropped" label. Themes are matched across calls by name or shared keywords, so one topic can land on two rows. Here, Q2's "Investment banking and sales & trading" (17 mentions) sits on its own row, which is why markets shows a gap in Q2. The Q2 cell for "AI and operating leverage" is Q2's "Revenue and earnings growth" theme, joined through the keyword "operating leverage".
The settlement theme has no such ambiguity. It's new, and it's tagged risk.
Themes are summaries. Word counts are checkable. "How often did BAC say 'settlement' and 'operating leverage' on its last four calls?" GetTranscriptPhraseTrend, once per phrase:
| Call | "settlement" | "operating leverage" | per 10k words |
|---|---|---|---|
| FY2025 Q4 (14 Jan) | 0 | 16 | 17.5 |
| FY2026 Q1 (15 Apr) | 0 | 22 | 18.1 |
| FY2026 Q2 (14 Jul) | 0 | 27 | 29.9 |
| FY2026 Q3 (14 Oct) | 11 | 13 | 9.8 |
The rate per 10,000 words matters because calls differ in length. The quotes it returns make the comparison in two lines:
And the settlement, from Q3: "we've adjusted those results to take into account the non-tax deductible regulatory settlement charge of $1.2 billion, which is a $0.17 per share impact."
"Show BAC's daily prices from 9 to 20 October." GetStockPrices:
Across companies I compare phrase counts rather than tone scores: a count means the same thing on every transcript. "How often did JPMorgan and Bank of America say 'agentic' on their last four calls?"
| Call | JPM | BAC |
|---|---|---|
| FY2025 Q4 | 0 | 0 |
| FY2026 Q1 | 0 | 1 |
| FY2026 Q2 | 0 | 1 |
| FY2026 Q3 | 7 (7.2 per 10k) | 3 (2.3 per 10k) |
The word went from absent to present at both banks in the same quarter. The Q3 quotes returned for both are analyst questions, one at BAC asking about "agentic AI as potentially driving more yield-seeking behavior, the potential risk of flight for more idle consumer deposits".
Using Equibles, for <TICKER>:
1. Compare tone, hedging and themes of the last two earnings calls.
2. Show four quarters of theme history; for each new or dropped theme,
check how it was paired before calling it new.
3. Count the two phrases that moved most, per 10,000 words, with quotes.
4. Add the daily closes around the latest call.
5. Count one phrase across <TICKER> and <PEER> for the same quarters.
The scored calls and transcripts are also on each company's calls page, like BAC earnings calls and JPM earnings calls.
Nothing here is investment advice.